Model comparison
Qwen3.5-2B-Base vs Qwen2.5-VL-7B-Instruct
At Q4_K_M, Qwen3.5-2B-Base is the smaller download — 1,274,396,032 bytes against 4,683,072,032. At long context the gap widens: Qwen3.5-2B-Base's KV cache at 32K is 4.7× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Qwen3.5-2B-Base | Qwen2.5-VL-7B-Instruct | |
|---|---|---|
| Parameters | 2.3B | 8.3B |
| Architecture | qwen35 | qwen2vl |
| Layers | 24 | 28 |
| Native context | 262,144 | 128,000 |
| Mixture of experts | no | no |
| Quantizations published | 8 | 50 |
| Smallest quantization | 1.15 GiB | 1.93 GiB |
| Q4_K_M | 1.19 GiB | 4.36 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Qwen3.5-2B-Base | Qwen2.5-VL-7B-Instruct | Ratio |
|---|---|---|---|
| 4,096 | 0.05 GiB | 0.22 GiB | 4.66× |
| 8,192 | 0.09 GiB | 0.44 GiB | 4.66× |
| 16,384 | 0.19 GiB | 0.88 GiB | 4.67× |
| 32,768 | 0.38 GiB | 1.75 GiB | 4.67× |
| 65,536 | 0.75 GiB | 3.50 GiB | 4.67× |
| 131,072 | 1.50 GiB | 7.00 GiB | 4.67× |